The notion of Trustworthy AI (TAI) has become increasingly pivotal in academic and public discussions on the safe and ethically acceptable development and deployment of AI systems. Most notably, it represents the conceptual foundation of the European effort to provide an ethics-based regulation for AI. However, the notion of TAI is not free from problems, and its very conceptual tenability has been called into question. Different authors have argued that trustworthiness depends on features, such as the trustee’s motivations and adherence to moral obligations, that AI systems clearly lack, making the notion of TAI a categorical error. This talk aims to provide a defense of the notion of TAI by arguing that not only there are ways to legitimately ascribe trustworthiness to AI systems, but also that doing so brings significant advantages. As a matter of fact, unlike notions such as responsible or reliable AI, the one of TAI allows us to effectively capture aspects related to AI systems’ accuracy and robustness and at the same time account for the value-laden dimensions of their deployment.
Join at: imt.lu/aula1